Adsorption of Selected Pharmaceuticals and an Endocrine Disrupting Compound by Granular Activated Carbon. 1. Adsorption Capacity and Kinetics

Transcription

1 Environ. Sci. Technol. 2009, 43, Downloaded by MICHIGAN STATE UNIV on September 19, Adsorption of Selected Pharmaceuticals and an Endocrine Disrupting Compound by Granular Activated Carbon. 1. Adsorption Capacity and Kinetics ZIRUI YU, SIGRID PELDSZUS,* AND PETER M. HUCK NSERC Chair in Water Treatment, Department of Civil and Environmental Engineering, University of Waterloo, 200 University Avenue West, Waterloo, Ontario, Canada, N2L 3G1 Received July 15, Revised manuscript received October 22, Accepted December 16, The adsorption of two representative PhACs (naproxen and carbamazepine) and one EDC (nonylphenol) were evaluated on two granular activated carbons (GAC). The primary objective was to investigate preloading effects by natural organic matter (NOM) on adsorption capacity and kinetics under conditions andconcentrations(i.e., ng/l) relevantfordrinkingwatertreatment. Isotherms demonstrated that all compounds were significantly negatively impacted by NOM fouling. Adsorption capacity reduction was most severe for the acidic naproxen, followed by the neutral carbamazepine and then the more hydrophobic nonylphenol. The GAC with the wider pore size distribution had considerably greater NOM loading, resulting in lower adsorption capacity. Different patterns for the change in Freundlich K F and 1/n with time revealed different competitive mechanisms for the different compounds. Mass transport coefficients determined by short fixed-bed (SFB) tests with virgin and preloaded GAC demonstrated that film diffusion primarily controls mass transfer on virgin and preloaded carbon. Naproxen suffered the greatest deteriorative effect on kinetic parameters due to preloading, followed by carbamazepine, and then nonylphenol. A type of surface NOM/biofilm, which appeared toaddanadditionalmasstransferresistancelayerandthusreduce film diffusion, was observed. In addition, electrostatic interactions between NOM/biofilm and the investigated compounds are proposed to contribute to the reduction of film diffusion. A companion paper building on this work describes treatability studies in pilot-scale GAC adsorbers and the effectiveness of a selected fixed-bed model. Introduction * Corresponding author fax: ; speldszu@ uwaterloo.ca. Pharmaceutically active compounds (PhACs) and endocrine disrupting compounds (EDCs) have been detected in both raw waters and finished drinking water albeit at trace levels (1-3). Mixtures of EDCs and PhACs at nanogram per liter levels have the potential to induce adverse effects in human cell lines (4), and the removal of these compounds through common drinking water treatment processes is therefore an emerging concern. Granular activated carbon (GAC) adsorbers are effective for removing conventional micropollutants such as pesticides. It was therefore proposed that GAC adsorption processes would be promising for the removal of PhACs and EDCs (5). Although studies have evaluated the adsorption on activated carbon of a broad range of PhACs and EDCs (2, 6, 7), adsorption was not always effective (6, 7). The observed diversity in removal efficiencies is attributable to the diverse physicochemical properties of the contaminants. Detailed adsorption characteristics are however not available, except for some isotherms for a limited number of PhACs and EDCs (8, 9) using µg/l, or even higher equilibrium concentrations, well above drinking water sourcewater concentrations (usually low µg/l to ng/l (10)). Since isotherm (e.g., Freundlich) parameters are dependent on the concentration range (11) and adsorption kinetics have not been published yet for PhACs and EDCs on GAC, performance prediction and further design of GAC adsorbers for these compounds is impeded. For conventional micropollutants, both GAC adsorption capacity and rates are substantially reduced due to preloading from background natural organic matter (NOM), leading to a much shortened bed service life (11-14). The preloading effects can be primarily attributed to competition for adsorption sites and pore blockage or constriction. At low NOM loading, competition, both reversible and irreversible, is the main contributor to reduced capacity. At longer run times, reductions in both adsorption capacity and rate are predominantly attributable to pore blockage or constriction (13, 15). However, in practice, several mechanisms may act simultaneously. The inherent diversity and much lower concentrations of PhACs and EDCs may lead to fouling phenomena different from those for conventional micropollutants, necessitating investigations such as those described herein. This study investigated the preloading effect from background NOM on adsorption capacity and kinetics for two PhACs (naproxen and carbamazepine) and one EDC (nonylphenol) at environmental levels on selected GACs. Understanding fouling characteristics and mechanisms provides the basis for the application of GAC adsorber models for process design, described in a companion paper (16). Materials and Methods Target Compounds. Two PhACs (i.e., naproxen and carbamazepine), and one EDC (nonylphenol) were chosen following a comprehensive evaluation process (3). Although the three compounds have similar molecular weights ( g/mol), their log K OW values range from 2.5 for carbamazepine, 3.18 for naproxen to 5.8 for nonylphenol, and their pk a values range from 4.2 for naproxen to 10.3 for nonylphenol. Note that the ph corrected log K OW for naproxen is at a ph of 7.5. Further details are given in Supporting Information (SI) S1. Adsorbents. Coal-based Calgon Filtrasorb 400 (F400) (Pittsburgh, PA) and coconut shell-based PICA CTIF TE (CTIF) (Columbus, OH) were selected (Table 1 in the SI). GAC particle densities were determined according to Sontheimer et al. (11). The Brunauer-Emmett-Teller (BET) surface areas (SA) and pore volumes were determined using a Micromeritics surface area and porosimetry analyzer (SAPA 2020). CTIF has a greater percentage of micropores (<20 Å) and a larger BET SA than F400, for both the majority of the pore volume is <20 Å. Waters. Both ultrapure water and post-sedimentation (PS) water from a treatment plant were used. Details are provided in the SI /es801961y CCC: $ American Chemical Society VOL. 43, NO. 5, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY Published on Web 02/06/2009

2 Preloading. The PS water was directed to two sand filters and then to preloading columns (filled with original size virgin carbons) having an i.d. of 2.54 cm (1 in.) with a hydraulic loading of approximately 6 m/h. The preloaded carbon samples (various times up to 16 weeks) were freeze-dried for at least 48 h, and subsequently sieved to obtain the U.S. mesh fraction for isotherm tests. The remaining carbon was used in the SFB tests, and previous sieve analysis demonstrated that removal of this small fraction would not substantially change its mean particle size. All preloaded carbon was stored in a desiccator prior to use. Isotherms. Isotherms using preloaded carbon were obtained using the bottle point method for each compound individually at an initial concentration of approximately 1 µg/l (18, 23). The actual spiked concentration was determined by GC/MS (3). Uncrushed U.S. mesh carbon was used since crushing may open pores blocked by background NOM and lead to an overestimation of adsorption capacity (13, 17). Preliminarily tests indicated apparent equilibrium, for all practical purposes, after 21 days. For other experimental details see ref 18. Adsorption capacities are expressed by the Freundlich equation (eq 1). q e ) K F C e 1 n where q e is the solid phase equilibrium solute concentration; C e is the liquid phase equilibrium solute concentration; K F and 1/n are constants. The Freundlich equation was chosen based on previous work (18). Short Fixed-Bed (SFB) Tests. The SFB approach (19) was applied for determining kinetic parameters. The uncrushed GAC (diameter 1 mm) was directly packed into the glass column (i.d cm) between two layers of glass beads after being soaked for 24 h in deionized water. SFB depth was increased with increasing preloading times. The feed solution contained the three target compounds at a constant concentration (approximately 500 ng/l each) with a hydraulic loading of 6 m/h. Actual concentrations were monitored in the influent and effluent (3). Long-term (380 h) and shortterm (27-30 h) SFB tests were conducted for virgin and preloaded carbons, respectively. The breakthrough curves were used for determining external and internal diffusion coefficients using nonlinear regression of the pore and surface diffusion model (PSDM). Mathematical Model and Parameterization. The PSDM (11-13, 20) was used to describe adsorption at low concentrations. This model includes external film diffusion (quantified as β L ) and subsequent pore transport either via pore diffusion (indexed as pore diffusion coefficient D p or impedance τ p, D p ) D L /τ p, D L is the free diffusivity), or by surface diffusion (indexed as surface diffusion coefficient D s ). Attachment of the solute at the adsorption site is described by eq 1. The detailed PSDM equations have been discussed elsewhere (11, 21). In this research the numerical solution for the PSDM was based on orthogonal collocation (21). β L values first estimated using the Gnielinski correlation (11) were used as initial values for nonlinear regression to determine β L using SFB data and the PSDM. The SFB breakthrough data were used to calibrate actual film and internal diffusion coefficients (either D s or D p ) as described in the SI. Results and Discussion Fouling Effect on Adsorption Capacity. In our previously reported investigations of F400 carbon in ultrapure water and with competitive co-adsorption (no preloading effect) from background NOM (18), naproxen and carbamazepine generally had isotherms comparable to 2-methylisoborneol (1) FIGURE 1. Effect of preloading time on Freundlich parameters for (a) naproxen, (b) carbamazepine, and (c) nonylphenol. (MIB) and geosmin, but much lower adsorption affinities than atrazine at equilibrium concentrations lower than 1 µg/l. Nonylphenol was least adsorbable in the investigated concentration range (18). In the present study, isotherms for naproxen and carbamazepine were determined on F400 and CTIF carbon preloaded for 1, 3, 5, 8, and 16 weeks. As expected, the impacts from preloading were severe (Figure 1a-c, and Tables 2 and 3 in the SI). K F values dropped substantially within the first 5-8 weeks. After 16 weeks, the K F values for naproxen were less than 0.5% of their virgin carbon values. For carbamazepine, K F on F400 after 16 weeks was only 1.2% of the virgin carbon value, whereas on CTIF it was approximately 30%. Since K F and 1/n are highly correlated, the adsorption capacities after 16 weeks preloading were examined at two liquid phase equilibrium concentrations (500 and 50 ng/l) (Figure 2 in the SI). For naproxen, approximately 2-12% of the capacity remained, compared to 16-27% for carbamazepine. In contrast, Knappe et al. (13) reported that the remaining atrazine capacity on a wood-based GAC decreased to approximately 50% (as measured by K F ) of the virgin carbon value after five months preloading with river water. Similarly, 40-60% capacity reduction (K F at C e ) 0.31 mg/l) was observed for trichloroethene (TCE) on GAC preloaded with ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 5, 2009

3 TABLE 1. Kinetic Parameters Determined on Virgin GACs from Long-Time SFB Tests a naproxen carbamazepine nonylphenol β L ( 10-3 cm/s) τ β L ( 10-3 cm/s) τ β L ( 10-3 cm/s) D s ( cm 2 /s) F ( ) b 1 (0.77 >4) 3.04 ( ) 0.99 (0.52 >4) 5.99 ( ) 3.24 ( ) CTIF 3.06 ( ) 1 (0.53 >4) 3.79 ( ) 1.50 ( ) 5.47 ( ) 3.74 ( ) a Film and pore diffusion model used for naproxen and carbamazepine, film and surface diffusion model used for nonylphenol. b 95% confidence intervals. Downloaded by MICHIGAN STATE UNIV on September 19, a river water for 25 weeks (22). The much lower (several orders of magnitude) applied concentrations in the present study would likely lead to a reduced ability for more hydrophilic compounds such as naproxen and carbamazepine to compete with the partially preloaded NOM or access the effective adsorption sites. The difference may also be partly attributable to the different water matrices and carbons used. For nonylphenol, a much lower K F did not significantly decrease after the initial drop within the first five weeks (Figure 1c) and the adsorption capacity was only reduced by approximately 11-58% after 16 weeks (Figure 2 in the SI). The above-mentioned concentration effect seems therefore to be much less pronounced for the hydrophobic nonylphenol. We previously found that virgin F400 had a higher capacity for naproxen and similar capacities for carbamazepine and nonylphenol compared to virgin CTIF (18). Figure 2 in the SI shows that over time F400 is more vulnerable than CTIF carbon to capacity reduction from fouling, which is in agreement with its higher DOC loading in the preloading system (23). F400 has a higher pore volume for pores larger than 8 Å, pores accessible to typical NOM molecules. A positive correlation therefore apparently exists between the severity of fouling and the percentage of secondary microand macropores in a long-term preloaded carbon. However, some batch studies (24, 25) showed that a greater percentage of secondary micropores would reduce the competition between NOM and micropollutants (e.g., atrazine) which also adsorb primarily in this size range, i.e., 8-20 Å. Therefore, naproxen and carbamazepine, which are comparable to atrazine in structure and size, should resist competition on F400 with its wider pore size distribution. However, only in the first 3 weeks of preloading did F400 have a higher capacity for naproxen and carbamazepine (Figure 1a and b); at longer times, the capacity of F400 decreases more rapidly. Therefore, the carbon with the wider pore size distribution seems subject to more fouling after prolonged preloading because more adsorption sites can be easily accessed by a larger fraction of the NOM, leading to greater site occupancy and more severe pore blockage. A higher secondary micropore volume loss for F400 carbon in Table 1 in the SI further confirms this. Interpretations of Different Capacity Reduction Patterns. For naproxen, K F greatly decreased before the 5th week and continued to decrease slowly until the 16th week, whereas 1/n remained relatively constant for the first three weeks and then increased, finally approaching unity (Figure 1a). K F for carbamazepine was not substantially reduced during the first week, and then continuously decreased thereafter (Figure 1b); 1/n increased only slightly or essentially remained constant throughout the entire period. For nonylphenol (Figure 1c), K F remained almost constant after the first five weeks; correspondingly, 1/n increased, approaching unity. These observations, to some extent, reflect two contradictory patterns in the literature. As for carbamazepine, a group of almost parallel isotherms for TCE on GAC preloaded for various times were found (11). Almost constant 1/n was also observed for atrazine on preloaded carbon (13). Others, however (17, 26), observed an increase in 1/n for TCE with preloading time, as observed here for naproxen. These conflicting results suggest that it is very difficult to generalize concerning competitive effects, which include direct competition, non-desorbable preloading, and pore blockage/ constriction. The contribution of each mechanism is likely dependent on the specific adsorbate-adsorbent pair. Two mechanisms might contribute to the changing pattern of Freundlich parameters for naproxen. In the early preloading stage, only very low molecular weight NOM can rapidly diffuse to the adsorption sites that naproxen molecules can access equally well, thus only leading to a reduction of K F. Another explanation for the early stage phenomena would be complete pore blockage by adsorption of lowmolecular-weight NOM, resulting in a loss of capacity without changing the overall adsorption site energy distribution (17). Both mechanisms may contribute to a loss in BET surface area prior to five weeks (Table 1 in the SI). At longer preloading times, the simultaneous increase in 1/n and decrease in K F could be caused by strongly adsorbable NOM which is tightly bound to high energy sites, thus changing the heterogeneity of the carbon (17). Meanwhile, large-molecular-weight externally deposited NOM may slowly diffuse into the carbon, possibly partially blocking access to adsorption sites. This type of NOM adsorption may occur over a longer period, and may not significantly reduce BET surface area (Table 1 in the SI). Therefore, BET surface area may not be a good indicator of effectiveness for adsorbing micropollutants, especially after long preloading times. For carbamazepine, only a minor change in 1/n on preloaded F400 and a constant 1/n on CTIF suggest that the heterogeneity of the carbon was not substantially impacted by preloading. For virgin carbon, we found that carbamazepine is more strongly adsorbable than naproxen on the two GACs in the investigated concentration range (18). Hence, carbamazepine should be able to compete for more adsorption sites occupied by NOM. This would lead to more parallel isotherms on GAC preloaded for a prolonged time. The mechanisms deemed responsible for the decreasing capacity for carbamazepine on preloaded carbon are both direct competition with desorbable NOM and pore blockage. Nonylphenol is a more hydrophobic, long straight chain molecule (Figure 1 in the SI). A small cross sectional area ( 5 Å) makes it more accessible to the primary micropores. As summarized by others (24), the majority of NOM cannot access primary micropores. Therefore, the overall capacity reduction of nonylphenol due to direct competition should be limited. Another possible mechanism may relate to nonylphenol s high log K OW, giving it a greater tendency to partition onto the deposited NOM. As reported by Carter et al. (17), this hypothesis is based on the observation that our 1/n values for nonylphenol are almost unity. Therefore, as preloading increases, the organic matrix formed by adsorbed NOM would not influence nonylphenol s partitioning. Adsorption Kinetics on Virgin Carbon. Long-term SFB tests were performed to determine the kinetic parameters (Table 1) for virgin carbons. The surface diffusion model (SDM) including both surface and film diffusion better fit the breakthrough data for nonylphenol (Figure 3 in the SI). In contrast, the kinetic parameters for naproxen and car- VOL. 43, NO. 5, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

4 FIGURE 2. PSDM sensitivity analysis for naproxen (a), carbamazepine (b), and nonylphenol (c) breakthrough in virgin F400 SFB reactor. bamazepine were better fit using the pore diffusion model (PDM). As expected, impedance (τ) values for naproxen and carbamazepine are all close to 1, suggesting that the pores in virgin GAC are open for the diffusion of these two compounds. Surface diffusion coefficients (D s ) for nonylphenol were of the magnitude expected, similar to the D s value for atrazine obtained under similar experimental conditions (13). The experimentally determined β L values are 1-3 times larger than the values estimated using the Gnielinski correlation (23). This is likely partially attributable to deviations from the particle sphericity assumed in the correlation (11). Sensitivity analyses assessed the relative contribution of isotherm and kinetic parameters to the breakthrough profiles (Figure 2). The baseline utilized the parameters previously determined on virgin GAC. Generally, the parameters were varied by 50%, however, the 1/n value for nonylphenol was only changed 25% because it approached unity, and τ by 100% and to infinity, to highlight the impact of this parameter. Results for CTIF carbon were similar to those for F400 (23). The predicted profiles are most sensitive to K F. A decrease in K F led to much faster breakthrough, suggesting that a small capacity reduction due to fouling could have a great impact. An increase in 1/n, implying the loss of high energy adsorption sites, accelerates breakthrough, though less strongly. It was originally expected that the influence of film diffusion would only be pronounced during early breakthrough (11). Surprisingly, the simulations were more sensitive to β L than to τ for naproxen and carbamazepine (Figure 2a and b). This was further confirmed by applying the extreme condition -τ at infinity. These observations suggest that, at very low concentrations, film diffusion predominantly controls mass transport. Furthermore, this likely contributed to the difficulties in accurately determining the pore diffusion coefficient. A 50% increase in β L for nonylphenol raised its baseline breakthrough profile in an almost parallel way (Figure 2c). Surface diffusion, although having less effect, seems to influence the entire nonylphenol breakthrough profile. Figure 2c implies that nonylphenol mass transport is controlled by both film and surface diffusion, especially in the later stages. The greater influence of film diffusion on naproxen and carbamazepine breakthrough agrees with results obtained by others for MIB at ng/l levels (14). That study attributed a similar observation of slow film diffusion to the low flowrate used. However, our flowrate was approximately 6 m/h, within the range of full-scale practice. Therefore, the rate-limiting effect of film diffusion can be attributed to the low concentrations used. For very low solute concentrations, the capacity of the carbon is very large. Thus it is not unreasonable to suppose that the small amounts of solutes that reach the outer surface are very quickly transported into the granule (27). Therefore, the major mass transport resistance exists between the liquid-solid interface and the bulk solution. Fouling Effect on Kinetics. Film diffusion coefficients for all three compounds on both carbons generally decreased rapidly during the first five weeks and then slowly thereafter (Table 4 in the SI). The general trend coincided with DOC breakthrough (not shown) for both carbons, which were almost exhausted for DOC within five weeks. This suggests that the decrease in film diffusion was attributable to the rapid adsorption of NOM on the GAC external surface in the early stages. Slow adsorption of NOM after five weeks could partially contribute to the slow decrease in β L values generally observed during that period. The observed decreases in β L are in accordance with several previous studies on TCE (12) and atrazine (13). Carter and Weber (12) proposed two main mechanisms: first, that the accumulation of NOM on the outer carbon surface increases the viscosity and thickness of the mass transfer film, thus decreasing diffusion rates and fluxes; and second, that the adsorbed NOM forms a film covering the outer surface, resulting in a reduced effective external surface area. However, the present study can only infer that the accumulation of NOM introduces another resistance layer because the kinetic tests could not distinguish between transport through the hydrodynamic boundary layer and through the newly formed NOM layer. Instead, the reduced film diffusion due to NOM film formation is incorporated into the experimentally determined β L value. Although it may be more accurate to describe the β L values for preloaded carbon as apparent film diffusion coefficients, the term film diffusion coefficient or β L value was retained for consistency. Figure 3 demonstrates that a NOM film (possibly with biological contributions) partially or even totally blocked the surface pore openings on preloaded carbon, especially for CTIF. The carbons roughness is reduced, which may decrease hydrodynamic perturbations, leading to lower film diffusion coefficients. The impact of a given preloading time on film diffusion depends on the specific compound (Table 4 in the SI). For virgin carbons, naproxen has similar β L values to carbam ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 5, 2009

5 Downloaded by MICHIGAN STATE UNIV on September 19, FIGURE 3. Comparison of SEM images between virgin and 16-week preloaded GAC. azepine (Table 1). However, βl for naproxen decayed much faster on preloaded carbon. βl for nonylphenol was least impacted, and its 16-week values were still substantially larger than the Gnielinski values for virgin carbon. The different reduction rates in βl for naproxen and carbamazepine may be attributed to the changed electrostatic interaction between adsorbents and adsorbates. Fairey et al. (28) provided evidence that phzpc dropped from on virgin F400 to on carbon preloaded with lake water. This is because typical surface water NOM has carboxylic acid groups and is therefore negatively charged at typical drinking water ph values (29). In these ph ranges, naproxen is dissociated and would be most strongly repelled by the negatively charged carbon surface. In contrast, electrostatic interactions would be much weaker for neutral carbamazepine. In addition, even small amounts of adsorbed NOM can lead to a dramatic change in surface charge (26, 30). Therefore, it can be expected that electrostatic repulsion could quickly increase, accounting for the more rapid decay of film diffusion for a negatively charged solute (e.g., naproxen) during early preloading. Another possible reason for the reduction in βl for naproxen is the adsorption or growth of bacteria on the carbon surface. The formation of bacterial colonies decreases the volume of the largest macropores and lowers the phzpc, thus possibly retarding film transport. This mechanism was confirmed using Escherichia coli (31). In our study, heterotrophic plate counts (HPCs) on freeze-dried 16-week preloaded CTIF carbon confirmed limited presence of bacteria. However, the contribution of bacteria on the carbon surface to a decrease in film diffusion needs further investigation. Table 5 in the SI presents the internal diffusion coefficients for preloaded GAC. Some of the estimated τ values for naproxen and carbamazepine are much larger than expected. For example, the τ values of 48 and 69 are considered unrealistic. In summarizing results, Sontheimer et al. (11) stated that the maximum τ value should be 4 after a long period of operation. Our regression analyses showed that the PDM lost virtually all sensitivity to τ, as discussed later. For nonylphenol, the surface diffusion rate on F400 decreased rapidy after the third week (Table 5 in the SI), TABLE 2. Required Impedance to Trigger Mass Transport Predominantly Controlled by Pore Diffusion on F400 Carbon naproxen carbamazepine preloading time (wks) τ τ in accordance with the trend reported for atrazine adsorption on F400 (32). In contrast, on CTIF, the Ds value decreased sharply within the first three weeks, and then reached a steady state. The observed absorbent-specific decreasing pattern for Ds may be due to the different carbon pore structures. Impact of Film Diffusion Decay on Mass Transport. To interpret the loss of sensitivity in model calibrations for τ, the Biot number for the PDM was employed (eq 2) for filmpore diffusion. Bip ) βldp 2Dp (2) in which, Bip is the pore diffusion Biot number (ratio of external liquid phase mass transfer rate to pore diffusion rate); dp is GAC particle diameter. The Biot number should be greater than for pore diffusion to predominantly control mass transport (11). If the actual τ is larger than the τ value in Table 2 (calculated using eq 2), mass transport will be substantially controlled by pore diffusion. The required τ values on virgin F400 carbon are for both naproxen and carbamazepine. However, in general, the pore diffusion coefficient of a compound for virgin carbon should be close to its free diffusivity, i.e., a τ value near unity. This suggests that mass transfer for naproxen and carbamazepine on virgin F400 would be largely limited by film diffusion which is consistent with Figure 2a and b. Furthermore, the required τ for naproxen VOL. 43, NO. 5, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

6 on F400 greatly increases between 0 and 16 weeks due to a significant increase in apparent film diffusion resistance. An extremely large required τ (18-36) was obtained for naproxen on 16-week preloaded F400 carbon, suggesting that it is not possible for pore diffusion alone to significantly control mass transport in this case. Therefore, accurate estimation of film diffusion for naproxen on preloaded carbon is important for accurate breakthrough prediction. For carbamazepine, the expected required τ increases from on virgin carbon to 6-12 on sixteen-week preloaded carbon, implying the possibility of pore diffusion to significantly influence mass transfer if the carbon is extensively preloaded (Table 5 in the SI). Practical Implications. In this study the preloading effects on both adsorption capacities and mass transfer were significantly influenced by the solutes characteristics. Different capacity reduction patterns indicate that mechanisms by which preloading exerts an impact cannot easily be generalized. In general, the impact of preloading can be expected to be more severe for hydrophilic or dissociated solutes such as naproxen. Since the number of PhACs and EDCs is extremely large, investigations of preloading mechanisms should be undertaken for compounds grouped by similar physicochemical properties to aid in judging the appropriateness of GAC for a particular compound. The preloading effects for GAC removing PhACs and EDCs may be much stronger than for pesticides because the concentrations are typically much lower. This study suggested that the GAC with a wide pore size distribution had considerably greater NOM loading after prolonged time, leading to lower capacities. It can therefore be inferred that a GAC with a large fraction of pores in the size range of the target PhAC or EDC and a small fraction of larger pores would be preferred. This study demonstrated that film diffusion is the primary controlling mass transfer mechanism for very low concentrations of contaminants, and therefore must be accurately determined for good breakthrough predictions. Many studies, including ours, have provided evidence that film diffusion is difficult to predict with a correlation (e.g., Gnielinski) due to the complexity of GAC particle topography and system hydrodynamics. This study showed that the difference between the experimentally determined β L and the correlated value is compound-specific. Therefore, it is strongly recommended that β L be experimentally determined under similar hydrodynamic conditions to those expected in practice. In addition, any pretreatment options implemented to reduce high molecular weight NOM prior to GAC adsorbers would be beneficial. Nevertheless, the mechanisms for film diffusion decay require further research. Acknowledgments We acknowledge PICA USA Inc. and Calgon Carbon Corporation for their donation of carbon, and the Natural Sciences and Engineering Research Council of Canada and our Industrial Research Chair partners ( ca/watertreatment/) for funding. Supporting Information Available Additional information including five tables, four figures, and additional references are available free of charge via the Internet at Literature Cited (1) Servos, M. R.; Smith, M.; McInnis, R.; Burnison, K.; Lee, B-H.; Seto, P.; Backus, S. Presence and removal of acidic drugs in drinking water in Ontario, Canada. Water Qual. Res. J. Can. 2007, 42 (2), (2) Stackelberg, P. E.; Gibbs, J.; Furlong, E. T.; Meyer, M. T.; Zaugg, S. D.; Lippincott, R. L. Efficiency of conventional drinking-watertreatment processes in removal of pharmaceuticals and other organic compounds. Sci. Total Environ. 2007, 377 (2-3), (3) Yu,Z.;Peldszus,S.;Huck,P.M.Optimizinggaschromatographicmass spectrometric analysis of selected pharmaceuticals and endocrine-disrupting substances in water using factorial experimental design. J. Chromatogr. A. 2007, 1148, (4) Pomati, F.; Castiglioni, S.; Zuccato, E.; Fanelli, R.; Vigetti, D.; Rossetti, C.; Calamari, D. Effects of a complex mixture of therapeutic drugs at environmental levels on human embryonic cells. Environ. Sci. Technol. 2006, 40 (7), (5) Weyer, P. and Riley, D. Endocrine Disruptors and Pharmaceuticals in Drinking Water; AWWA Research Foundation and American Water Works Association: Denver, CO, (6) Westerhoff, P.; Yoon, Y.; Snyder, S.; Wert, E. Fate of endocrine disruptor, pharmaceutical, and personal care product chemicals during simulated drinking water treatment processes. Environ. Sci. Technol. 2005, 39 (17), (7) Snyder, S. A.; Adham, S.; Redding, A. M.; Cannon, F. S.; DeCarolis, J.; Oppenheimer, J.; Wert, E. C.; Yoon, Y. Role of membranes and activated carbon in the removal of endocrine disruptors and pharmaceuticals. Desalination 2007, 202, (8) Ternes, T. A.; Meisenheimer, M.; McDowell, D.; Sacher, F.; Brauch, H.; Haist-Gulde, B.; Preuss, G.; Wilme, U.; Zulei-Seibert, N. Removal of pharmaceuticals during drinking water treatment. Environ. Sci. Technol. 2002, 36 (17), (9) Choi, K. J.; Kim, S. G.; Kim, C. W.; Kim, S. H. Effects of activated carbon types and service life on removal of endocrine disrupting chemicals: Amitrol, nonylphenol, and bisphenol-a. Chemosphere 2005, 58 (11), (10) Kolpin, D.; Furlong, E.; Meyer, M. T.; Thurman, E. M.; Zaugg, S. D.; Barber, L. B.; Buxton, H. T. Pharmaceuticals, hormones, and other organic wastewater contaminants in U.S. streams, : A national reconnaissance. Environ. Sci. Technol. 2002, 36 (6), (11) Sontheimer, H., Crittenden, J. C., Summers, R. S. Activated Carbon for Water Treatment, 2nd ed.;dvgw-forschungsstelle: Karlsruhe, Germany, (12) Carter, M. C.; Weber, W. J. Modeling adsorption of TCE by activated carbon preloaded by background organic matter. Environ. Sci. Technol. 1994, 28, (13) Knappe, D. R. U.; Snoeyink, V. L.; Roche, P.; Prados, M. J.; Bourbigot, M-M. Atrazine removal by preloaded GAC. J. Am. Water Works Assoc. 1999, 91 (10), (14) Gillogly, T. E. T. MIB adsorption in drinking water treatment. Ph.D. dissertation, University of Illinois, Urbana, Illinois, (15) Kilduff, J.; Wigton, A. Sorption of TCE by humic-preloaded activated carbon: Incorporating size-exclusion and pore blockage phenomena in a competitive adsorption model. Environ. Sci. Technol. 1999, 33, (16) Yu, Z.; Peldszus, S.; Huck, P. M. Adsorption of selected pharmaceuticals and an endocrine disrupting compound by granular activated carbon. 2. Model prediction Environ. Sci. Technol. 2009, 43, (17) Carter, M. C.; Weber, W. J.; Olmstead, K. P. Effect of background dissolved organic matter on TCE adsorption by GAC. J. Am. Water Works Assoc. 1992, 84, (18) Yu, Z.; Peldszus, S.; Huck, P. M. Adsorption characteristics of selected pharmaceuticals and endocrine disrupting compoundnaproxen, carbamazepine and nonylphenol-on activated carbon. Water Res. 2008, 42 (12), (19) Weber, W. J.; Liu, K. T. Determination of mass transport parameters for fixed-bed adsorbers. Chem. Eng. Commun. 1980, 6, (20) Jarvie, M. E.; Hand, D. W.; Bhuvendralingam, S.; Crittenden, J. C.; Hokanson, D. R. Simulating the performance of fixed-bed granular activated carbon adsorbers: Removal of synthetic organic chemicals in the presence of background organic matter. Water Res. 2005, 39, (21) Crittenden, J. C.; Hutzler, N. J.; Geyer, D. G. Transport of organic compounds with saturated groundwater flow: Model development and parameter sensitivity. Water Resour. Res. 1986, 22 (3), (22) Summers, R. C.; Haist, B.; Koehler, J.; Ritz, J.; Zimmer, G.; Sontheimer, H. The influence of background organic matter on GAC adsorption. J. Am. Water Works Assoc. 1989, 81 (5), (23) Yu, Z. Analysis of selected pharmaceuticals and endocrine disrupting compounds and their removal by granular acti ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 5, 2009

7 vated carbon in drinking water treatment. Ph.D. dissertation, University of Waterloo, Waterloo, ON, Canada, (24) Karanfil, T.; Dastgheib, S. A.; Mauldin, D. Exploring molecular sieve capabilities of activated carbon fibers to reduce the impact of NOM preloading on trichloroethylene adsorption. Environ. Sci. Technol. 2006, 40, (25) Pelekani, C.; Snoeyink, V. L. Competitive adsorption between atrazine and methylene blue on activated carbon: The importance of pore size distribution. Carbon. 2000, 38, (26) Kilduff, J. E.; Karanfil, T.; Weber, W. J. TCE adsorption by GAC preloaded with humic substances. J. Am. Water Works Assoc. 1998, 90 (5), (27) Wolborska, A. External film control of the fixed bed adsorption. Chem. Eng. J. 1999, 73, (28) Fairey, J. L.; Speitel, G. E.; Katz, L. E. Impact of natural organic matter on monochloramine reduction by granular activated carbon: The role of porosity and electrostatic surface properties. Environ. Sci. Technol. 2006, 40, (29) Newcombe, G.; Drikas, M.; Assemi, S.; Beckett, R. Influence of characterized natural organic material on activated carbon adsorption: I Characterisation of concentrated reservoir water. Water Res. 1997, 31 (5), (30) Morris, G.; Newcombe, G. Granular activated carbon: The variation of surface properties with the adsorption of humic substances. J. Colloid Interface Sci. 1993, 159 (2), (31) Rivera-Utrilla, J.; Bautista-Toledo, I.; Ferro-Garcia, M. A.; Moreno-Castilla, C. Bioadsorption of Pb(II), Cd(II), and Cr(VI) on activated carbon from aqueous solutions. Carbon. 2003, 41, (32) Schideman, L. C.; Marinas, B. J.; Snoeyink, V. L.; Campos, C. Three-component competitive adsorption model for fixed-bed and moving-bed granular activated carbon adsorbers. Part II. Model parameterization and verification. Environ. Sci. Technol. 2006, 40, ES801961Y VOL. 43, NO. 5, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY